LLM Tracing

LLM Tracing opens on Conversations: every conversation your agents held in the window — voice calls, web chat, SMS, email and the rest — with what it cost, how long it ran, its turns and its errors. Totals split voice from text, each channel shows its typical and expensive conversation, a chart stacks spend per day by channel, and Outliers pulls out the conversations that cost or ran far more than their channel usually does, saying why.

The Turns tab is the flight recorder underneath. Each row is one turn — which model answered, how long it took, how many tokens it used, and whether it errored — coloured by duration so a slow or failed turn surfaces without being sorted for.

Cost is shown per turn where we can establish it — read off the call itself, or matched to its gateway spend record. A turn that went another route reads "not measured" rather than showing you a zero, and the footer tells you how many of the turns on screen are covered. Token counts arrive the same way and can be missing on the same turns.